This project asks why disinformation on social media is so successful in the realm of political speech. By analyzing the twitter accounts of fringe actors claiming to report political “news”, we explore the underlying cultural narratives that make this type of information meaningful and appealing. Do accounts that purposefully spread false information (disinformation) use the language of openness, democracy and truth, even as they muddy the political discourse on the internet? What other cultural queues do these accounts use to signal followers sympathetic to their political views? And finally, what topics generate the most engagement and popularity?*
We construct a novel dataset of tweets pertaining to politically polarizing Canadian news events mined from Twitter using neural relevance matching. Each tweet is annotated along several dimensions (e.g., “Is this tweet satirical?”, “Is this tweet factually accurate?”, “Did the author put a lot of effort into writing this tweet?”, etc.). We present an analysis of the dataset, exploring the differences between malicious and innocuous discourse along these dimensions, and propose a method for integrating these factors with deep learning methods to improve the detection of malicious content.
*Description provided by Prof. Komeili.